2023-10-08 03:42:35 +00:00
|
|
|
% CSCI 5521 Introduction to Machine Learning
|
|
|
|
% Rui Kuang
|
|
|
|
% Demonstration of Classification by 2-D Gaussians
|
|
|
|
|
2023-10-08 05:38:10 +00:00
|
|
|
clf;
|
|
|
|
prior1 = 0.3;
|
|
|
|
prior2 = 0.7;
|
|
|
|
|
2023-10-08 06:50:20 +00:00
|
|
|
% mvndis = @(X, mu, Sigma, prior) ( ...
|
|
|
|
% -1/2 * log(2*pi) ...
|
|
|
|
% - log(Sigma) ...
|
|
|
|
% - power(X - mu, 2) / (2 * power(Sigma, 2)) ...
|
|
|
|
% + log(prior) ...
|
|
|
|
% );
|
|
|
|
|
2023-10-08 03:42:35 +00:00
|
|
|
mu1 = [-1 -1];
|
|
|
|
mu2 = [1 1];
|
|
|
|
|
|
|
|
% Equal diagnoal covariance matrix
|
2023-10-08 17:48:13 +00:00
|
|
|
% Sigma1 = [1 0; 0 1];
|
|
|
|
% Sigma2 = [1 0; 0 1];
|
2023-10-08 03:42:35 +00:00
|
|
|
|
|
|
|
% Diagnoal covariance matrix
|
|
|
|
% Sigma1 = [1 0; 0 0.5];
|
|
|
|
% Sigma2 = [1 0; 0 0.5];
|
|
|
|
|
|
|
|
% Shared covariance matrix
|
|
|
|
% Sigma1 = [1 0.3; 0.3 0.5];
|
|
|
|
% Sigma2 = [1 0.3; 0.3 0.5];
|
|
|
|
|
|
|
|
x1 = -10:.1:10; x2 = -10:.1:10;
|
|
|
|
|
|
|
|
% covariance matrix (increase the range for visualization)
|
2023-10-08 17:48:13 +00:00
|
|
|
Sigma1 = [1 0.1; 0.1 0.5];
|
|
|
|
Sigma2 = [0.5 0.3; 0.3 1];
|
|
|
|
x1 = -40:.1:40; x2 = -40:.1:40;
|
2023-10-08 03:42:35 +00:00
|
|
|
|
|
|
|
[X1,X2] = meshgrid(x1,x2);
|
|
|
|
%pdf1
|
2023-10-08 06:50:20 +00:00
|
|
|
% F1 = mvnpdf([X1(:) X2(:)],mu1,Sigma1);
|
|
|
|
F1 = mvndis([X1(:) X2(:)], mu1, Sigma1, prior1);
|
2023-10-08 17:48:13 +00:00
|
|
|
F1 = reshape(F1,length(x2),length(x1));
|
2023-10-08 03:42:35 +00:00
|
|
|
subplot(1,2,1);
|
|
|
|
surf(x1,x2,F1); hold on;
|
|
|
|
|
|
|
|
%pdf2
|
2023-10-08 06:50:20 +00:00
|
|
|
% F2 = mvnpdf([X1(:) X2(:)],mu2,Sigma2);
|
|
|
|
F2 = mvndis([X1(:) X2(:)], mu2, Sigma2, prior2);
|
2023-10-08 17:48:13 +00:00
|
|
|
F2 = reshape(F2,length(x2),length(x1));
|
2023-10-08 03:42:35 +00:00
|
|
|
surf(x1,x2,F2);
|
|
|
|
caxis([min(F2(:))-.5*range(F2(:)),max(F2(:))]);
|
|
|
|
axis([-4 4 -4 4 0 .4])
|
|
|
|
xlabel('x1'); ylabel('x2'); zlabel('Probability Density');
|
|
|
|
|
|
|
|
%decosopm boundary
|
|
|
|
%F1 = mvnpdf([X1(:) X2(:)],mu1,Sigma1);
|
|
|
|
%F1 = reshape(F1,length(x2),length(x1));
|
|
|
|
%F2 = mvnpdf([X1(:) X2(:)],mu2,Sigma2);
|
|
|
|
%F2 = reshape(F2,length(x2),length(x1));
|
|
|
|
|
|
|
|
cmp = F1 > F2;
|
|
|
|
subplot(1,2,2);
|
|
|
|
imagesc(X1(:),X2(:),cmp);
|
|
|
|
|
|
|
|
xlabel('x1'); ylabel('x2');
|
2023-10-08 06:50:20 +00:00
|
|
|
|
|
|
|
|
|
|
|
function res = mvndis(X, mu, Sigma, prior)
|
2023-10-08 17:48:13 +00:00
|
|
|
[len, d] = size(X);
|
2023-10-08 06:50:20 +00:00
|
|
|
res = zeros(len, 1);
|
|
|
|
|
|
|
|
for i = 1:len
|
|
|
|
x = X(i,:);
|
|
|
|
mdist = (x - mu) * inv(Sigma) * (x - mu).';
|
2023-10-08 17:48:13 +00:00
|
|
|
res(i) = -d/2*log(2*pi) - 1/2*log(det(Sigma)) - 1/2*mdist + log(prior);
|
2023-10-08 06:50:20 +00:00
|
|
|
end
|
|
|
|
% 1 x 2
|
|
|
|
% (1 x 2) x ((2 x 2) x (2 x 1))
|
|
|
|
% X - mu = 40401 x 2
|
|
|
|
% (40401 x 2) x (2 x 2) x (2 x 40401)
|
|
|
|
% mdist = (X - mu) * inv(Sigma) * (X - mu).';
|
|
|
|
% res = -log(2*pi) - 1/2*log(det(Sigma)) - 1/2*mdist + log(prior)
|
|
|
|
% res = zeros(size(X));
|
|
|
|
% [l1, l2] = size(X);
|
|
|
|
%
|
|
|
|
% for i1 = 1:l1
|
|
|
|
% for i2 = 1:l2
|
|
|
|
% cell = -log(2*pi) - 1/2*log(det(Sigma))
|
|
|
|
% res(i1, i2) = cell
|
|
|
|
% end
|
|
|
|
% end
|
|
|
|
end
|